logistic regression

[#1]

hello.
my question is how can I calculate the probability of each data sample belonging to a specific class in logistic regression?

[#2 · Accepted solution]

Hi @maryam_nourmand,

Welcome in the Community !

If you're using Logistic Regression, you have the possibility to save probability formulas for your classes : Logistic Platform Options (jmp.com)
Simply click on red triangle, and then on "Save Probability Formula" :

New columns will be added in your datatable, with probabilities for each class for every rows of your datatable, as well as Most Likely Class (default threshold is 0.5) : 

Example here is from Titanic Passengers dataset, available in menu "Help", "Sample Index", "Exploratory Modeling", Titanic Passengers.

I hope this will answer your question,

[#3]

tanx for your responding.
if i want use robust logistic regression what should i do?

[#4]

Hi @maryam_nourmand,

What do you mean by "robust" logistic regression ?

In JMP Pro, you have several penalized estimation methods for the Generalized logistic regression : Lasso, Ridge, Elastic Net ... :

Is it what you're looking for ?

[#5]

in Nominal Logistic Fit part , i couldnt find any method for estimate parameters..
my intention with robustness is to ensure that the estimation of logistic regression model parameters is not influenced by outliers

[#6]

I think the Response Screening (jmp.com) platform might be what you're looking for ?

It enables to reduce the sensitivity of tests to outliers.

[#7]

tanx

[#8]

Run your model (https://www.jmp.com/support/help/en/18.0/#page/jmp/logistic-regression-models.shtml#) and then save the probability formula to your table 

[#9]

I would add two things to Victor and jthi's responses.  First, when you save the probability formula, the way in which the probabilities are calculated from the log of the odds can be seen by the formulas in those probability columns.  This is the link between the response variable, which is actually the log of the odds rather than the discrete response variable, and the estimated probabilities.  Second, the "most likely" prediction is based on which probability is greater - in other words, it uses a 50% probability cutoff for making predictions.  That is rarely the best cutoff in application, particularly because the costs associated with false positive and false negative predictions are rarely symmetric.  When you run the logistic regression, you can find "Decision Threshold" under the red arrow which allows you to explore different probability cutoffs and the resulting classifications.